Dr M. S. Khuroo, PO Box 3354, MBC 46, Riyadh 11211, Saudi Arabia. E-mail: firstname.lastname@example.org
Background : The treatment effects of primary prophylactic endoscopic variceal ligation are unclear.
Aim : To compare the treatment effects of endoscopic variceal ligation and β-blockers for primary prophylaxis of oesophageal variceal bleeding. In addition, a subgroup analysis was done with the purpose to delineate differences in the effects of intervention that were biologically based.
Methods : We performed a literature search for randomized controlled trials, which compared the treatment effects of endoscopic variceal ligation with β-blockers for primary prophylaxis of oesophageal variceal bleeding. Of the 955 articles screened, eight randomized-controlled trials including 596 subjects (285 with endoscopic variceal ligation and 311 with β-blockers) were analysed. Outcomes measures evaluated were first gastrointestinal bleed, first variceal bleed, all-cause deaths, bleed-related deaths and severe adverse events. The measure of association employed was relative risk; with heterogeneity and sensitivity analyses.
Results : Variceal obliteration was obtained in 261 (91.6%) patients and target β-blockers therapy was achieved in 294 (94.5%) patients (P = 0.19). Endoscopic variceal ligation compared with β-blockers significantly reduced rates of first gastrointestinal bleed by 31% (RR, 0.69; 95% CI: 0.49–0.96; P = 0.03; NNTB = 15) and first variceal bleed by 43% (RR, 0.57; 95% CI: 0.38–0.85; P = 0.0067; NNTB = 11). All-cause deaths and bleed-related deaths were unaffected (RR, 1.03; 95% CI: 0.79–1.36; P = 0.81 and RR, 0.84; 95% CI: 0.44–1.61; P = 0.60 respectively). Severe adverse events were significantly less in endoscopic variceal ligation compared with β-blockers (RR, 0.34; 95% CI: 0.17–0.69; P = 0.0024; NNTB = 28). Sensitivity analysis of five trials published in peer review journals and four trials with high quality showed results similar to those seen in the primary analysis of all the eight trials, confirming stability of conclusions. Following variceal obliteration with endoscopic variceal ligation, oesophageal varices recurred in 83 (29.1%) patients. Seven (28.1%) patients bled with one fatal outcome. In subgroup analyses, endoscopic variceal ligation had significant advantage compared wtih β-blockers in trials including ≤30% patients with alcoholic cirrhosis, >30% patients with Child Class C cirrhosis and >50% patients with large varices.
Conclusions : In patients with cirrhosis with moderate to large varices and who have not bled, endoscopic varices ligation compared with β-blockers significantly reduced bleeding episodes and severe adverse events, but had no effect on mortality.
In patients with cirrhosis, the prevalence of oesophageal varices is 60% and the risk of gastrointestinal bleeding is approximately 30%. High risk factors for first haemorrhage include the severity of cirrhosis, the size of oesophageal varices and the presence of red signs on varices. About 30–50% of patients admitted for the first episode of variceal bleeding die within 6 weeks. Accordingly, 10–15% of untreated and unselected patients with cirrhosis die from a first episode of gastrointestinal bleeding.1–3 These findings strongly suggest that prophylaxis should be implemented to prevent first gastrointestinal bleeding, especially in patients with a high risk of haemorrhage.
Non-selective β-adrenergic antagonists namely propranolol and nadolol, are widely used in the prevention of first bleeding in patients with cirrhosis and portal hypertension.4 These drugs reduce the relative risk (RR) of bleeding by approximately 45% and the mortality rate by 20% at 2 years.5 However, β-blockers (BB) have several disadvantages for long-term therapy. About over one-third of patients do not exhibit decrease in portal pressure despite adequate β-blockade.6 This is important as protection afforded by BB depends on the degree of portal pressure reduction.7 Other problems include several contraindications, frequent adverse events necessitating withdrawal of therapy, need for long-term therapy and the risk of rebound bleeding upon abrupt cessation of therapy.4 Transjugular intrahepatic portosystemic shunts (TIPS) and endoscopic sclerotherapy are more effective than drug therapy but at the price of higher side-effects and, in some studies, higher mortality.8, 9
Because of its efficacy and safety, endoscopic variceal ligation (EVL) has become the method of choice for secondary prophylaxis of variceal bleeding.10 As a result of this favourable experience, there has been interest in extending use of EVL to the primary prophylaxis of oesophageal variceal bleeding. A meta-analysis of four randomized trials comparing EVL with BB for primary prevention of variceal bleeding revealed inconclusive results.11
The purpose of this study was to perform a meta-analysis of randomized-controlled trials (RCT) comparing EVL with BB for primary prevention of variceal bleeding and determine the effect of EVL on first gastrointestinal bleed, all-cause deaths, bleed-related deaths and severe adverse events. In addition, a subgroup analysis was done with the purpose to delineate differences in the effects of intervention that were biologically based.
We sought to conduct a meta-analysis that compares the treatment effects of EVL and BB on the specific end-points in patients with portal hypertension and oesophageal varices and no prior episode of gastrointestinal bleed. A protocol was written which specified several aspects of the meta-analysis as per defined guidelines.12 A checklist was developed for purpose of data entry.
Acquisition of data
‘Medline’ (National Library of Medicine, Bethesda, MD, USA) and ‘EMBASE’ (Elsevier, New York, NY, USA) were searched from 1966 to July 2004 to locate published research in the area of oesophageal variceal haemorrhage. MeSH terms used for key and text word searching included ‘liver cirrhosis’, ‘gastrointestinal haemorrhage’, ‘portal hypertension’, ‘oesophageal and gastric varices’, ‘ligation’, ‘adrenergic β-antagonists’, ‘propranolol and nadolol’. We also searched the Cochrane Registry of controlled trials database. Published abstracts corresponding to the American Gastroenterology Association, the American College of Gastroenterology, the American Society of Gastrointestinal Endoscopy, United European Gastroenterology Week meetings and the British Society of Gastroenterology were also reviewed. Finally, a manual search was performed using bibliographies from each full-published report.
Criteria for selection. The following selection criteria were applied: (i) study design: prospective RCT; (ii) language of publication: both English and other languages; (iii) study population: patients with portal hypertension and oesophageal varices and no prior episode of gastrointestinal haemorrhage; (iv) intervention: oesophageal variceal ligation (EVL) and (v) comparison intervention: non-selective β-adrenergic blocker therapy (BB). Trials, which employed EVL for secondary prophylaxis or compared EVL with no treatment, nitrates and sclerotherapy or combined EVL with drug treatment in the intervention group were excluded. Reporting of duplicated studies was excluded by examining the author list, parent institution, sample size and results.
Outcome measures. We employed five outcome measures to compare the treatment effects of EVL and BB. These included: (i) first gastrointestinal bleed; (ii) first variceal bleed; (iii) all-cause deaths; (iv) bleed-related deaths and (v) severe adverse events.
Definitions. Terminology and definitions of adverse events was based on Baveno III consensus report.13 All outcome measures were reported from the time patient was randomized up to the end of follow-up. Gastrointestinal bleeding was defined as overt bleeding from upper gastrointestinal tract presenting as haematemesis and/or melaena. Bleeding from oesophageal varices was diagnosed if upper gastrointestinal endoscopy revealed active bleeding or clot on oesophageal varices or endoscopic signs of recent variceal bleeding and no other visible mucosal bleeding lesion. Bleeding was considered to be caused by oesophageal ulcers as a result of band ligation if there was active bleeding or an adherent clot on the oesophageal ulcer. All-cause deaths were deaths from any cause. Bleed-related deaths were defined as death within 6 weeks of an episode of gastrointestinal bleeding. Severe adverse events were those which led to the discontinuation of BB treatment (persistent bradycardia, bronchospasm, cardiac failure, etc.) or were life-threatening (oesophageal perforation; ligation ulcer bleeding) or caused significant morbidity (oesophageal stricture). Variceal obliteration was defined by the complete absence of varices or presence of varices grade 1 only. Recurrent varices were defined as the presence of varices grade II or more or varices with a diameter of at least 3 mm after initial successful obliteration. Withdrawals were defined as patients excluded after randomization or lost to follow-up or excluded from the study because of non-compliance. To minimize a type 1 error and to be sure not to find a difference when none exists, all analyses was performed according to the intention-to-treat method.14
Three (MSK, KLCF and SD) investigators independently evaluated trials for inclusion and for outcome measures and subsequently resolved any disagreement by discussion.
Assessment of study quality
Two reviewers (MSK and YSK) independently assessed trial quality by examining the allocation sequence generation, allocation concealment, investigator blindness, completeness of follow-up and efficacy of randomization. Each trial was scored as per established criteria (maximum score 10).15 Trials were considered high quality if methodological quality score was 5 or more. Any differences between reviewers were resolved by consensus.
The allocation sequence was classified as adequate if based on computer generated random numbers, table of random numbers or similar. The allocation concealment was classified as adequate if the allocation sequence was concealed until the moment of randomization by a central independent unit, sealed envelopes or similar. Investigator blindness involved use of identical placebo in the control arm and having outcome assessors and patients unaware of who received which study intervention. We defined complete follow-up as 90% or more with no preferential loss to follow-up in one group over the other. Randomization was adequate if pre-treatment prognostic variables in the intervention and control arm were depicted in tabular form and revealed no statistically significant differences.16, 17
Three of us (MSK, KLCF and SD) read independently each reference for quantitative analysis and a checklist of questions was completed. The quantitative data abstracted included sample size; patient characteristics; procedure details for EVL; protocol for BB therapy; duration of follow-up and evaluation of outcome measures listed above. Discrepancies of analysis were resolved by consensus.
The measure of association used in this meta-analysis was RR with 95% confidence interval (CI). Summary RR with 95% CI was calculated using fixed-effect (inverse variance-weighted method) and random-effect models (Der Simonian and Laird).18, 19 A statistically significant result was assumed when the 95% CI did not include 1. Relative risk reduction (RRR) was calculated as the proportional reduction in the rates of bad events in EVL over BB groups and accompanied by 95% CI. Relative risk increase (RRI) was calculated as the proportional increase in the rates of bad events in EVL over BB group and accompanied by 95% CI. When the RR for an outcome was significant, the number needed to treat to either benefit (NNTB) or harm (NNTH) one additional patient with 95% CI was calculated for that measure.16
Statistical heterogeneity between trials was evaluated by the Cochran chi-square test and was considered to exist when P < 0.1. Where heterogeneity was detected; accepted methods of exploration of statistical heterogeneity using clinical parameters were used.14 In the absence of statistical significant heterogeneity, only the RR by the fixed-effect model is given in the results, whereas both the fixed and the random model are presented in the case of statistically significant heterogeneity.
Publication bias was investigated through visual inspection of funnel plots whereby odds ratios were plotted against sample size. Because graphic evaluation can be subjective, a rank correlation test for publication bias was employed.20, 21 We performed cumulative meta-analysis as a measure to assess the robustness of the point estimate of effect size over time and to identify influence of individual interventions. For this studies were sequentially summarized by adding one study at a time in the order of year of publication.14
We performed several sensitivity analyses to assess the stability of conclusions to assumptions about the probabilities used in the analysis and were done as an assessment of the methodology. For this purpose, we assessed whether type of publication, and quality of trials could influence the results of our meta-analysis. Next, we performed subgroup analysis with the purpose to delineate differences in the effects of intervention that were biologically based. These included the influence of following factors on the outcome measures namely: proportion of alcoholic cirrhosis in the trials (≤30% vs. >30%); proportion of cirrhosis with Child's Class C in the trials (≤30% vs. >30%) and proportion of patients with large oesophageal varices in the trials (≤50% vs. >50%). Subgroup analysis on the dose of BB could not be done, as four of the five trials used mean dose of 60–77 mg/day (low dose) and only one trial used mean dose of 120 mg/day (high dose).
All analyses and calculations were performed using a program (easyma) developed by the Department of Clinical Pharmacology of the University hospital in Lyon, France.22
We identified 955 articles for review. Eight trials met criteria for inclusion.23–30 The reasons for excluding articles (followed by the number excluded) were: not a randomized trial (744), different intervention on comparison (121), different patient population (41), reference not related to study objective (37) and evaluation of different outcome (12). There was unanimity between the authors about selection of relevant articles.
Of the eight selected trials, five had been published as full manuscripts in peer-reviewed journals24, 25, 28–30 and three were published only in an abstract form (Table 1).23, 26, 27 All patients included in five trials had cirrhosis of liver,25, 26, 28–30 while majority (92%) of the patients had cirrhosis of liver in another trial.24 In two trials published as abstracts cause of portal hypertension was not known.23, 27 Sample size calculations had been done in three trials.28–30 In one trial an interim analysis of the sample size projected that even more than 200 patients in either arm would be far from sufficient to demonstrate difference in bleeding rate.30 The study was stopped after a total of 152 were recruited in the trial. Mean duration of follow-up was <1 year in one trial,26 1–2 years in six trials23–25, 27–29 and more than 2 years in one trial.30 Two trials were multicentre based. The trial by Lui et al.28 involved six centres and had three treatment arms. Only two centres had sufficient experience in EVL to randomize patients to all three arms. Remaining centres randomized patients to the two drug arms (propranolol and isosorbide-5-mononitrate). The treatment arm involving nitrates was excluded from analysis. The trial by Schepke et al.30 involved 27 centres and of these only six centres enrolled 10 patients or more each (total 100). All trials fulfilled the inclusion criteria for meta-analysis. In addition, two trials performed hepatic haemodynamic studies and included patients with hepatic venous pressure gradient of ≥12 mmHg.25, 27 All five peer-reviewed trials defined exclusion criteria namely age limits (<18 or >75 years), liver cancer, severe co-morbid conditions and contraindications to BB therapy.24, 25, 28–30 In four trials, 106 (21.9%) of the 483 eligible patients fulfilling inclusion criteria were excluded on such criteria.24, 25, 29, 30
Table 1. Study design and methodological quality of trials
Type of publication
Follow-up duration (months)
Study quality and score*
Allocation sequence generation (score)
Allocation concealment (score)
Follow-up, % (score)
–, data not known or evaluated; *, none of the trials was blinded, due to nature of interventions, so investigator blindness was not scored.
Multiband every week, mean sessions – 2 ± 1.2 per patient, obliteration in 69 of 75 patients
Propanolol on incremental basis to reach target heart rate, mean dose – 77 mg/day, target heart rate in 61 of 77 patients
34.4 ± 18.9 (0.1–73.1)
Computer generated block randomization for each centre (2)
Central coordinating unit (3)
Tabulated, comparable (2)
Four peer-reviewed trials had adequate allocation sequence generation24, 28–30 and three trials had adequate allocation concealment.28–30 None of the trials was blinded due to nature of interventions. Randomization was adequate in all peer-reviewed trials.24, 25, 28–30 Four trials reported on completeness of follow-up and was over 95% of patients in all.24, 28–30 Four trials had methodological quality score of 5 or more.24, 28–30
A total of 596 subjects (285 in EVL arm and 311 in BB) were included in the trials (Table 2). The mean number of subjects per trial was 74 with range from 24 to 152. Mean age of patients was ≤50 years in three trials24, 25, 27 and >50 years in three trials.28–30 In two trials, age of the patients could not be ascertained.22, 25 Males formed the predominant (60–90%) patient population in all six trials who reported on sex distribution.24, 25, 27–30 Alcoholic cirrhosis constituted <30% of patient population in four trials23–25, 28 and >30% in another two trials.28, 30 Similarly, Child Class C liver cirrhosis constituted ≤30% in three trials25, 29, 30 and >30% in another three trials.24, 26, 28 Large varices (grade III or IV) were seen in ≤50% patients in two trials28, 29 and >50% in three trials.24, 25, 30 Only three trials reported on red signs on varices28–30 and were seen in around one-third patients in two trials. Associated gastric varices were reported in four trials24, 28–30 and were seen in 5–20% of patients. Agreement between investigators for the quality and quantitative scoring assessment was 100%.
Table 2. Patient characteristics in included trials
Four trials employed multiband variceal ligator;23, 25, 29, 30 one trial used single band ligator with an overtube to facilitate repeated insertion of the endoscope24 and one trial used single band ligator initially and multiband ligator in the latter part of study.28 Type of ligator used in two trials was not known.26, 27 EVL was repeated every week (two trials)24, 30 or every other week (three trials)23, 25, 28 or every 3–4 weeks (one trial)29 till varices were obliterated. Number of sessions needed to obliterate varices in different trials varied from mean of 2–3.4 per patient. Variceal obliteration was obtained in 261 (91.6%) patients. Variceal obliteration was obtained in all patients in four trials (98 patients).23–25, 27 Varices could not be obliterated in 25 patients in remaining four trials (187 patients).26, 28–30 Following variceal obliteration, endoscopy was performed at 1–3 months intervals to check for recurrent varices. During the follow-up period, 83 (29.1%; 95% CI: 3.9–23.5) patients developed recurrent varices. The variceal recurrence in various trials varied from 20 to 75%. Occurrence of bleeding from recurrent varices was recorded in four trials (32 patients)24, 25, 28, 29 and had occurred in seven (28.1%) patients with one fatal outcome. Recurrent varices were obliterated with fewer EVL sessions (mean 1.5 per patient). One trial reported second recurrence in 11 (14.7%) patients.30
Propranolol was the non-selective BB used in seven trials23–28, 30 and nadolol in one trial.29 BB were administered on incremental basis until predefined clinical target (25% decrease in baseline heart rate in seven trials23–27, 29, 30 and dose of propranolol 160 mg/day in one trial28) was achieved. Mean dose of BB used in different trials varied from 60 to 120 mg/day. Four trials used a low-dose BB (mean dose 60–77 mg/day)24, 25, 29, 30 and one trial employed a high-dose BB (median dose 120 mg/day).28 The dose of BB used in three trials was not known.23, 26, 27 Effective β-blockage (as per predefined targets) was achieved in 294 (94.5%) patients. During the study period, 53 (23.7%) patients were withdrawn from therapy. The reasons for withdrawals were: inability to continue therapy due to adverse events (39 patients), non-compliance to therapy (12 patients) and not known (2 patients).
Heterogeneity and publication bias of the trials
Heterogeneity (Cochran chi-square) of trials for all five outcome measures (first bleed, first variceal bleed, all-cause deaths, bleed-related deaths and severe adverse events) were not significant (P = 0.90, 0.74, 0.97, 0.91 and 0.71 respectively), suggesting that the trials for all end-points were homogeneous. Thus, all data for primary meta-analyses were reported as calculated by fixed-effect. Inspection of the funnel plots for all outcome measures did not reveal evidence of publication bias (rank test, P = 0.62, 0.65, 0.29, 0.34 and 0.65 respectively).
First gastrointestinal bleeding. All eight trials evaluated first gastrointestinal bleeding as an outcome measure (Figure 1). In the EVL group, 40 (14.3%; 95% CI: 8.5–20.1) patients had bled. The rate of bleeding in individual trials varied from 0 to 25.3%. The site of bleeding was from oesophageal varices in 28 patients, ligation-induced ulcers in eight, peptic ulcer in three and hypertensive gastropathy in one patient. Seven patients bled before the varices were obliterated. Recurrent varices bled in seven patients. In the BB group, 65 (20.9%; 95% CI: 13.9–27.8) patients bled. The bleeding rate in individual trials varied from 5.6 to 32.0%. The site of bleeding was from oesophageal varices in 58 patients, gastric varices in four and peptic ulcer in three patients.
Only one trial evaluated individually showed significant reduction in rate of bleeding in EVL compared with BB group (RR, 0.34; 95% CI: 0.12–0.94).24 Summary RR for all eight trials showed a significant reduction in rate of bleeding in EVL compared with BB group (RR, 0.69; 95% CI: 0.48–0.97; RRR, −31%; 95% CI: −51 to −3; P = 0.034; NNTB = 15; 95% CI: 8–18). A cumulative analysis revealed that the rate of bleeding between EVL and BB groups was not significantly significant till first five trials (size = 234) were sequentially added to each other (P = 0.065).23–27 EVL showed a significant reduction in rate of bleeding (P = 0.034) once trial of Lui et al.28 published in 2002 was added to the earlier five published trials (size = 344). This significant difference was stable once, other two trials published in 2004 were sequentially added to earlier six published trials (P = 0.014 and 0.031 respectively).29, 30
We also evaluated first variceal bleed as an outcome measure to compare the treatment effects of EVL and BB. Summary RR for all eight trials showed a significant reduction in rate of first variceal bleeding in EVL compared wtih BB group (RR, 0.57; 95% CI: 0.38–0.85; RRR, −43%; 95% CI: −62 to −15; P = 0.0067; NNTB = 11; 95% CI: 8–23). These results were similar to those obtained on first gastrointestinal haemorrhage as an outcome measure.
All-cause deaths. Seven trials evaluated all-cause deaths as an outcome measure (Figure 2).23–25, 27–30 In the EVL group, 66 (24.7%; 95% CI: 13.6–35.8) patients died; death rates in individual trials varied from 0 to 45.3%. The causes of death were as follows: variceal bleeding (12), bleeding from ligation-induced ulcer (two), hepatic failure (eight), sepsis (four), hepatorenal syndrome (three), hepatocellular carcinoma (two) and not mentioned (35). In the BB group, 71 (24.1%; 95% CI: 13.4–34.8) patients died; the death rates in individual trials varied from 0 to 42.8%. The causes of death were: gastrointestinal bleeding (19), hepatic failure (seven), sepsis (one), hepatorenal syndrome (two), hepatocellular carcinoma (one) and not mentioned (41).
None of the trials evaluated individually showed significant difference in rate of all-cause deaths between EVL and BB group. Summary RR for all seven trials showed no significant difference in death rates between EVL and BB group (RR, 1.03; 95% CI: 0.79–1.36; RRI, 3%; 95% CI: −21 to 36; P = 0.81). A cumulative analysis revealed no significant difference in death rates between EVL and BB group throughout the analyses.
Bleed-related deaths. Six trials evaluated bleed-related deaths as an outcome measure (Figure 3).23–25, 28–30 In the EVL group, 14 (5.5%; 95% CI: 1.5–9.5) patients had bleed-related deaths; rates in individual trials varied from 0 to 12.0%. In the BB group, 19 (6.7%; 95% CI: 3.2–10.2) patients had bleed-related deaths; rates in individual trials varied from 0 to 10.3%. None of the trials evaluated individually showed significant difference in rate of bleed-related deaths between EVL and BB group. Summary RR for all six trials showed no significant difference in death rates between EVL and BB group (RR, 0.84; 95% CI: 0.44–1.61; RRR, −16%; 95% CI: −56 to 61; P = 0.60). A cumulative analysis revealed no significant difference in bleed-related death rates between EVL and BB group throughout the analyses.
Severe adverse events. Six trials evaluated adverse events as an outcome measure (Figure 4, Table 3).24, 25, 27–30 Adverse events were reported in 103 (42.7%) patients in EVL group. The commonest adverse event was ligation-induced oesophageal ulcers, dysphagia and chest pain. However, majority of these events were of minor nature and did not cause any significant morbidity. Severe adverse events occurred in nine (3.7%; 95% CI: 1.63–5.77) patients; the rates in individual trials varied from 0 to 6.7%. This included ligation-induced oesophageal ulcer bleeds in eight patients with two fatal outcomes and overtube-induced oesophageal perforation in one patient.
Table 3. Adverse events in two intervention groups
* Had oesophageal ulcer bleed.
† Necessitated stopping of β-blockers.
‡ Nine (16.9%) of the 39 patients bled in the follow-up with two fatal outcomes.
Total (six trials/ 241 patients)
Total (six trials/ 264 patients)
Adverse events were reported in 148 (56.1%) patients in BB group. The commonest adverse events were hypotension, breathlessness and bradycardia. Severe adverse events necessitating withdrawal of BB occurred in 39 (14.7%; 95% CI: 4.68–24.72) patients; the rates in individual trials varied from 6.7 to 30.3%. Of these, nine (23.1%) patients bled on withdrawal of BB with two fatal outcomes.
Only one trial showed significantly lower severe adverse events in EVL over BB group (0.17; 95% CI: 0.04–0.62).28 Summary RR for all seven trials showed a significant reduction in rate of severe adverse events in EVL compared with BB group (RR, 0.34; 95% CI: 0.17–0.69; RRR, −66%; 95% CI: −83 to −32; P = 0.0024; NNTB = 28, 95% CI: 19–32). A cumulative analysis revealed no significant difference (P = 0.49) in severe adverse event rates till first three trials were sequentially added to each other (size = 143).24, 25, 27 EVL showed a significant reduction in rate of severe adverse events (P = 0.014) once trial of Lui et al.28 published in 2002 was added to the earlier three published trials. This significant difference was stable once other two trials published in 2004 were sequentially added to earlier four published trials (P = 0.0082 and 0.0024 respectively).29, 30
The study of the sensitivity analysis revealed that the summary of RR for all the end-points remained statistically stable (similar to summary RR obtained from eight trials) when meta-analyses were limited to five trials published in peer-review journals24, 25, 28–30 and four trials with high methodological quality score (Table 4).24, 28–30 The Cochran chi-square values for heterogeneity for all outcome measure were not significant, suggesting that the trials for these end-points were homogeneous.
Table 4. Sensitivity analyses of peer-reviewed and high quality trials
RR (95% CI)
P for heterogeneity
P for publication bias
EVL, endoscopic variceal ligation group; BB, β-blocker group; RR, relative risk; CI, confidence interval; NNTB, the number of patients who need to be treated to prevent one additional bad outcome in EVL group.
For subgroup analyses, all outcome measures were calculated, however, only first gastrointestinal bleeding revealed important differences between various subgroup and thus reported results are restricted to this outcome measure.
Proportion of alcoholic cirrhosis in the trials. In four trials24–26, 29 (size = 254) including ≤30% patients with alcoholic cirrhosis, EVL significantly reduced rate of first gastrointestinal bleeding (RR, 0.45; 95% CI: 0.21–0.95; P = 0.036). In contrast, EVL and BB revealed no significant difference in rate of first gastrointestinal bleeding in two trials28, 30 (size = 262) including >30% patients with alcoholic cirrhosis (RR, 0.63; 95% CI: 0.37–1.10; P = 0.085).
Proportion of cirrhosis with Child's Class C. In three trials25, 29, 30 (size = 282) including ≤30% patients with Child Class C cirrhosis, EVL and BB revealed no significant difference in rate of first gastrointestinal bleeding (RR, 0.67; 95% CI: 0.41–1.10; P = 0.11). In contrast, EVL significantly reduced rate of first gastrointestinal bleeding in another three trials24, 26, 28 (size = 234) including >30% patients with Child Class C cirrhosis (RR, 0.32; 95% CI: 0.13–0.77; P = 0.011).
Proportion of patients with large oesophageal varices. In two trials28, 29 (size = 210) including ≤50% with large varices, EVL and BB were equally effective in reducing rate of first gastrointestinal bleeding (RR, 0.53; 95% CI: 0.24–1.17; P = 0.21). In contrast, EVL significantly reduced rates of first gastrointestinal bleeding in three trials24, 25, 30 (size = 271) including >50% patients with large varices (RR, 0.57; 95% CI: 0.34–0.97; P = 0.04).
The purpose of this meta-analysis was to compare the efficacy of EVL with that of BB for multiple outcome measures in patients with portal hypertension and varices and who had no gastrointestinal bleed in the past. A meta-analysis on this subject published in 2001 had several limitations.11 The meta-analysis involved four randomized trials (size = 283) comparing EVL with BB and only two of these trials were published as peer-reviewed full papers (size = 119). EVL compared with BB reduced the risk of first variceal bleed, but had no effect on mortality.11 The number of included patients necessary to show a significant difference in survival rates between two groups was estimated to be approximately 500.3 Thus, firm conclusions could not be made from this analysis on the results obtained. The present analysis included eight trials and 596 patients and the results obtained, although similar had appropriate statistical power for the conclusions obtained. In addition, authors of the previous meta-analysis did not perform sensitivity analysis of trials comparing EVL with BB.11 Sensitivity analysis is essential to document stability of conclusions of primary meta-analysis and is based on analysis of data from high quality studies and or those published as peer-reviewed articles.14 Thus, conclusions that EVL was better than BB could not be firmly established on statistical basis due to lack of data on sensitivity analysis. We performed, in retrospect a sensitivity analysis on the data of the Imperiale and Chalasani's paper (RR, 0.45; 95% CI: 0.17–1.16; P = 0.099; P for heterogeneity = 0.16) and found that EVL compared with BB did not reduce rate of first variceal bleeding. This made the results of primary meta-analysis inconclusive. We agree this was due to small number of patients included. In contrast, we performed a sensitivity analysis on peer-reviewed and high quality trials separately in the present meta-analysis and found stability of conclusions of the primary meta-analysis in both models. The earlier meta-analysis, had also, failed to analyse adverse effects of two treatment regimens, which is so important especially for interventions used for prophylactic purposes.11 We analysed severe adverse events between two interventions and made some important observations in this regard.
We took several measures to make the results of this meta-analysis valid. Statistical heterogeneity of the trials was performed for each outcome measure and confirmed that the trials were homogeneous. In addition, trials were examined for clinical heterogeneity and found no obvious differences in study populations, interventions, or the way outcome measures were assessed. The analyses were performed by both fixed-effect and random-effect models and found no differences in the treatment effects. The random-effect model adjusts for among-trial variability by providing a more conservative estimate of a treatment effect by using wider confidence intervals. Sensitivity analyses of peer-reviewed articles and trials with high methodological quality score were performed for all outcome measures and the treatment effects determined resembled that of primary meta-analysis. Findings of cumulative analysis revealed the robustness of the point estimate of all-cause deaths and bleed-related deaths throughout the study period. For first gastrointestinal bleeds, first variceal bleeds and severe adverse events cumulative analysis results revealed stability once trials published in 2002 and later were sequentially added to earlier published trials. Another aspect, which ensures validity of meta-analysis, is completeness of literature search (publication bias). We made a comprehensive search of the literature as per defined protocol for meta-analysis and all relevant studies were included. Also, visual inspection and statistical evaluation of funnel plots for all outcome measures excluded such a possibility. On the basis of these data we believe the results of this meta-analysis are valid and addition of any future trials is unlikely to change the results of these outcome measures.
Assessing the methodological quality score for meta-analysis has been a subject of debate. Traditionally, trials are scored for individual methodological qualities and summary scores calculated.15 On the basis of summary scores, high quality trials are identified and separately analysed for sensitivity analysis. Recent data have shown that use of summary scores to identify high quality trials is problematic as these are not associated with treatment effects in RCT.31, 32 It was recommended to assess individual methodological aspects of each trial and their influence on effect size explored. For the present meta-analysis, we assessed individual methodological qualities of each trial as well calculated a summary quality score.
The EVL compared with BB reduced the rate of first gastrointestinal bleed by 31% with number needed to treat to benefit one patient as 15. Of the eight trials analysed for these outcome measures, only one trial assessed individually reduced rate of first bleed.23 An editorial on this trial commented that the result may have been due to exceptionally high bleeding rate (27.3%) in the BB group and/or lower mean dose (60 mg/day) of BB.2 However, rate of first bleed in BB group in trial by Schepke et al. was 25.3%, yet the latter trial failed to show reduced rate of bleeding in EVL group.30 In addition, the trial by Lui et al. was the only trial which had employed high dose (median dose: 120 mg/day) of BB,28 while the dose of BB in other three peer-reviewed published trials was similar (60–77 mg/day) to that employed in the trial by Sarin et al.24, 25, 29, 30
This meta-analysis showed that EVL compared with BB had no significant effect on all-cause deaths. We believe the lack of effect on all-cause deaths was due to overall small contribution of total deaths by gastrointestinal bleeds, the outcome effect significantly reduced by EVL. The deaths caused by bleeding contributed to only 34.2% of the total deaths and majority of deaths were caused by other complications of cirrhosis. It is obvious from these data that any benefit achieved in mortality through reduction in the bleed-related deaths (if any) would be insufficient to reduce the overall mortality.
The EVL compared with BB did not affect the bleed-related deaths. These data were intriguing as EVL significantly reduced incidence of first gastrointestinal bleeds. On further assessing the data we observed that gastrointestinal bleeding in EVL were fatal in 42.9% (12 of 28) of patients. In contrast, the first bleed in BB group was fatal in only 29.3% (17 of 58). Also, ligation-induced oesophageal ulcers bled in eight patients with two fatal outcomes. Increased severity and consequent mortality in gastrointestinal bleeds in EVL was possibly related to injury and ulcers caused by the ligation bands on varices. This was supported by the observations of all peer-reviewed trials that majority of the bleeds in EVL occurred in the period when varices were being eradicated, while the bleeds in BB group occurred throughout the follow-up period.
The reporting of severe adverse events is an important one. The severe adverse events were significantly lower in EVL compared with BB group. The nature of severe adverse events was quite different in the two interventions. The major problem in EVL group was ligation-induced oesophageal ulcer bleeds. Major concern was that two patients with variceal ulcer bleeds died. One patient had oesophageal perforation, which occurred with the use of an overtube in single band ligator. This is no more being used as EVL is now being done by multiband ligator. BB, because of their wide spread actions, caused a variety of adverse events. The most severe events leading to withdrawal were hypotension, bradycardia and breathlessness. There were no fatalities reported with BB. However, two patients who had withdrawn from BB had gastrointestinal bleeding on follow-up with fatal outcome.
Subgroup analysis revealed that EVL had significant advantage compared with BB in trials including ≤30% patients with alcoholic cirrhosis, >30% patients with Child Class C cirrhosis and >50% patients with large varices. The implications of these data need to be considered with caution, as relative efficacy of either intervention needs trials with a head-to-head comparison of differing biological parameters in the included patient populations. However, the data of this subgroup analysis may give us a clue to plan these studies in future.
What clinical lessons can be taken from this meta-analysis on the primary prophylaxis of oesophageal variceal bleeding? We believe EVL should be offered as first-line therapy in patients with cirrhosis of liver and large oesophageal varices who are non-compliant to BB as well as to patients in whom severe adverse events or contraindications preclude the use of BB (in all 23.7%). EVL should also be offered to patients who bleed on BB and include possibly those patients who show no adequate haemodynamic response to BB (20.9%). Given the very similar mortality rates in the two groups, BB therapy in remaining patients may be more convenient to use. Patients in both intervention groups need long-term therapy, continued drug treatment in BB group and repeated ligation in EVL group for recurrent varices and on this score the two interventions are balanced.
What are the limitations of this meta-analysis? As with all other meta-analysis, particular attention must be paid to the studies included. We included only prospective randomized-controlled studies. Three of the eight included trials were published as abstracts. All trials in this meta-analysis were open-labelled because of the nature of interventions. These issues may compromise the quality of the meta-analysis. However, the sensitivity analysis revealed that the summary RR for all the end-points remained statistically stable (similar to summary RR for five peer-reviewed and four high quality trials). Trials included in the meta-analysis were done in patients with different class of patients with cirrhosis, with varying dose of BB, and different protocol for EVL. However, trials revealed no statistical heterogeneity for all end-points and could be combined to obtain summary RR.
Quantification: MSK (40%), NSK (20%), KLCF (10%), YSK (10%), AAS (10%), SD (10%). Definition of work: Concept and protocol: MSK, NSK; acquisition of data: MSK, KLCF, SD; qualitative analysis of trials: MSK, YSK; quantitative analysis of trials: MSK, KLCF, SD; data analysis: MSK, AAS; statistical analysis: MSK, NSK; manuscript preparation: MSK, NSK; final approval: all authors.
The study was funded by a grant from Evidence Based Medicine (EBM) Unit of the Digestive Diseases Services, Srinagar, Kashmir, India. The authors are thankful to Mehnaaz Sultan Khuroo, MD, at Department of Pathology, Ramachandra Medical College and Research Institute, Porur, Chennai, India and to Saleem Kamili, PhD, at Division of Pathology, Center for Disease Control, Atlanta, Georgia, USA for help provided during this study.